This blog post was written by Stephan Lewandowsky and has been crossposted from the Psychonomic Society blog.
About 20 years ago a conference on working memory was held in Quebec City, Canada. One of the eminent visitors from the UK had to return home early. He successfully navigated to the Quebec City airport, flew to Toronto or Montreal to catch a connecting flight to Heathrow. All went well until a customs officer challenged his passport, pointing out that he did not look much like his (female) colleague who had shared the outbound flight and whose passport he was brandishing.
How could a gentleman check in for at least two flights and travel halfway around the world before discovering that his passport was actually hers? Check-in agents at major airports are certainly busy, but they are also very experienced—so how can they miss a rather obvious mis-match between an actual person and his (or her) passport photo?
A recent article in the Psychonomic Society’s journal Cognitive Research: Principles and Implications addressed the issue of expertise in ID photo recognition. This is one of a number of articles in a special issue dedicated to individual differences in face recognition edited by Vicki Bruce, Karen Lander, and Markus Bindemann. We have published two posts on this special issue already, here and here. Those posts provided a broad overview of the topic. In this post, I focus on the article by Megan Papesh that focused on the role of expertise and training.
Papesh chose to conduct her research with notaries, because notaries routinely establish the identity of their clients before witnessing their signatures on legal documents. Although highly routine and perhaps even boring, this task is far from trivial. There have been several cases involving high-stakes fraud in which impersonators were able to pass a notary’s identity check. For example, in one case criminals stole a murder victim’s driver’s license and hired an actor to impersonate him before the notary, thus obtaining access to the murdered man’s property.
More than 800 notaries participated in the study online. The notaries on average had more than 8 years’ experience on the job. (This was not reported in the paper but because the author put the data online at OSF I was able to compute this statistic in 6 minutes).
The participants completed 30 trials that required them to indicate via mouse click whether a photo of a person matched or did not match a photo ID. The figure below shows a sample stimulus:
Match and mismatch trials were randomly interspersed, and participants could not skip trials or go back to previous stimuli. After completing all 30 trials, the full stimulus set was presented along with the participant’s responses and the correct answers.
The main results for the notaries are shown in the next figure. The left column of panels presents accuracy data for the trials on which the face matched the accompanying ID, and the right column presents accuracy data for mismatch trials. In each column, accuracy (scaled to the range 0 to 1) is expressed as a function of various experience-related variables. The solid lines represent best-fitting regression lines.
It is clear that age negatively impacts performance (top row of panels), in line with much previous evidence that face-recognition abilities decline with advancing age. Strikingly, experience showed no relationship to accuracy: There was no improvement in performance across number of years of experience (bottom row), despite the wide range of that variable (from 2 to 48 years). Likewise, the number of times that the notaries were conducting ID-checks each week (center panels) also had no effect on their performance. The absence of those effects was confirmed by a Bayesian statistical analysis, which identified strong evidence for the null hypothesis.
A further analysis showed that the result was not due to a correlation between age and experience (older notaries tend to have more experience). The same pattern obtained for notaries from a single age bracket when performance was compared across levels of experience.
Papesh’s study also included a smaller sample of bank tellers and students. Bank tellers also need to check IDs routinely, whereas the students are unlikely to have any experience with this task. Thus, if experience matters then the students would be expected to perform worse than the two groups of experts.
There was no hint of any effect of expertise: Averaging across match and mismatch trials, the mean accuracy for notaries was 76%, compared to 76% for bank tellers and 79% for students. No, those slight differences were not statistically significant.
In addition to age, there was however one further strong effect: participants were more accurate on match trials than on mismatch trials. This can be seen in the above figure by comparing the intercept of the regression lines between the left and right columns. This difference is frequently observed in similar tasks, although the reasons for it remain unclear.
So why was there no effect of expertise?
The effects of expertise are pervasive: in an earlier post we noted that expert Sommeliers are so good at wine tasting that some have taken out £10,000,000 insurance on their olfactory system. Clearly, no one would bother insuring a random student’s olfactory system for that amount, so wine-tasting experience indubitably translates into enhanced performance. Similarly, we reported earlier that elite soccer referees perform better than their less elite colleagues. So if performance in domains as disparate as soccer and wine can benefit from experience, why is ID-checking an exception?
The answer likely involves the fact that experience need not translate into expertise. The professionals in Papesh’s study had plenty of experience but they were not trained in acquiring a particular skill. This explanation is supported by the finding that the only professional group that has ever been found to outperform novices on unfamiliar face matching involves forensic image examiners. Those examiners receive rigorous and extended training in facial image analysis, for example at the famous FBI Academy in Quantico, Virginia.
In case you are interested, there are now sites online that offer training in forensic facial analysis. (It may be advisable to await empirical validation of those offerings.)
For readers familiar with the literature on expertise, Papesh’s results should not be terribly surprising. After all, the idea that expertise is a result of deliberate practice rather than mere exposure and experience is a cornerstone of theories of expertise, as we have noted here before.